A new orthonormal wavelet packet decomposition for audio coding using frequency-varying modulated lapped transforms

Wavelet packet decompositions based on tree structured 2-channel filter banks with conjugate quadrature filters (CQF) have found many applications in the area of audio coding. Their time-frequency tiling is the dual of the time-varying modulated lapped transforms (MLT). We present a new orthonormal wavelet packet basis, which is constructed by the frequency-varying MLT. These can be viewed as the direct analogy to time-varying transforms. In contrast to the classical decomposition, the new transform shows a good bandpass behaviour without strong spectral sidelobes. Hence, it is particularly useful for the coding of audio signals. For the signals examined the coding gain is higher than that obtained with cascaded CQF filter banks. A fast algorithm for the new wavelet packet transform is possible with a polyphase implementation for the required modulated lapped transforms. This algorithm exceeds a fast CQF based wavelet packet decomposition, especially with the high frequency resolution, which is necessary in audio coding.